Abstract Accurate forecasting of petroleum production is essential for reservoir management, field development, and economic evaluation. Decline curve analysis (DCA) has been traditionally used to predict oil, gas, and water production. However, its limitations in handling complex reservoir behaviors often lead to less accurate predictions. This study examines the use of machine learning (ML) models to enhance production forecasting, comparing their performance to DCA with data from the Volve field. The study applied decline curve analyses (harmonic, exponential, and hyperbolic) and machine learning models to predict oil, water, and gas production, using parameters such as production date, down-hole pressure, temperature, and differential pressure tubing. The data was split 80/20% for training and testing, with Mean Absolute Error (MAE) and R-squared (R²) scores used for evaluation. DCA predicted R² scores of 42% for oil, 37% for gas, and −6.93% for water. In comparison, the Gradient Boosting Regressor achieved R² scores of 97.45% for oil, 97.26% for water, and 97.13% for gas. The Random Forest Regressor performed even better, with R² scores of 99.21% for oil, 98.23% for water, and 99.20% for gas. These results show that machine learning models, particularly Random Forest, significantly outperform traditional methods like DCA, offering more reliable and accurate production forecasts.
Dike et al. (Mon,) studied this question.